Comment by lnanek2

Since this is implemented by a neural network, maybe all you have to do to obtain that is train it with different voices.